TABLE OF CONTENTS

• Segmentation
• Adding Data and Dataset
• Segmentation Tools 
  • 1. Rectangle Tool: 
  • 2. Polygon Tool:
  • 3. SuperPix Tool: 
    • How to Use It 
  • 4. Smart Object Selection (SOS) Tool: 
    • How to Use It 
  • 5. Brush Tool:  
    • How to Use It 
  • 6. Eraser Tool:
    • How to Use It 

Segmentation

Image Segmentation is a process of assigning labels to a pixel or set of pixels with the same attributes in an image.

NeuralMarker offers various segmentation tools, to create pixel-wise mask annotations for each object in the image.

Adding Data and Dataset

Before using the Segmentation tools a user has to learn to do the following things:

1. To add dataset in NeuralMarker
2. To add data in NeuralMarker 

Select Category Type As "Segmentation" during adding of dataset.

Segmentation Tools 

1. Rectangle Tool: 

It's a drawing tool that helps to create a rectangular mask over the objects which have regular or non-complex boundaries in the image.

2. Polygon Tool:

It's a drawing tool that creates a polygon mask over the objects with irregular shapes.

To close the polygon mask the first point and last point of the mask should be the same.

3. SuperPix Tool: 

SuperPix is a Segmentation specific tool that produces multiple segments over the image by clustering pixels based on their colour similarity.

How to Use It 

Step 1: 

Select the Tool from the Toolbar adjacent to Canvas.

Step 2: 

A . Drag the SuperPix tool using the left mouse button to create the segmentation mask over the multiple segments of the object.

B. Segments size can be adjusted from "Finer" to "Coarser" by using the slider place next to the 

SuperPix tool.

As Calculation time for generating superpixel segments of various size increases with the image size. Therefore it is suggested to use images of less than 1 MB or of pixels size less than 2000x2000 pixels.

4. Smart Object Selection (SOS) Tool: 

A "Semi-Automatic Object Segmentation Tool" that generates a precise object segmentation mask on the objects in the image.

It accurately transforms the information of extreme points [left-most, right-most, top, bottom] of an object into a pixel-perfect segmentation mask with minimal user input.

How to Use It 

Step 1: 

Select the Tool from the Toolbar adjacent to Canvas.

Step 2:  Two-Click Approach

a. Draw the rectangle by clicking on the extreme left of the object.

b. Then sliding the cursor to the extreme right.

c. Click again to complete the rectangle over the object to create the segmentation mask.

After drawing the rectangle, the mask over the object will appear like the one shown below:

Few More Examples To demonstrate the capabilities of SOS Tool

A.

B.

5. Brush Tool:  

It's a free-flowing drawing tool that helps generate a circular and a planar mask over the object in the image. 

It can also be used in images where boundaries between objects and background are clearly visible.

The radius [in the range of 0 to 100 px ] of the brush can be changed using the slider next to the Brush tool.

How to Use It 

Step 1: 

Select the Brush Tool from the Toolbar adjacent to Canvas.

Step 2: 

Drag the brush using the left mouse button to create the segmentation mask over the object.

6. Eraser Tool:

It's an "over-segmentation correction tool" that helps to clear the unnecessary overlapping of segmentation masks between the object's boundaries.

The radius [in the range of 0 to 100 px ] of the eraser can be changed using the slider next to the Eraser tool.

How to Use It 

Step 1: 

Select the Eraser Tool from the Toolbar adjacent to Canvas.

Step 2: 

Drag the eraser using the left mouse button to remove the unnecessary mask between the objects.

Points to Remember

All the above displayed segmentation tool-sets will always assign a single category to a single pixel of an image.

In other words no single pixel of an image will be assigned multiple categories through the NeuralMarker Segmentation tools.

1. if an annotator draws an annotation which belong to a different category on an area in the image that overlap with the existing annotation of another category then the overlapped region between the two annotations is subtracted from that latest annotation.

2. Whereas, if an annotator draws an annotation on an area in the image that overlap with the existing annotation of the same category then that overlapped region of the recently created annotation is merged into the already existing annotation.